KR102001409B1 - Dynamic n-dimensional cubes for hosted analytics - Google Patents

Abstract

The online analysis processing system can include an n-dimensional cube structured using slice-based partitioning, where each slice comprises one or more hierarchies of data points. The regions of the hierarchy can be classified according to the computational demands associated with the regions. A scaling or replication mechanism can be applied to an area based on the computational demand associated with that area.

Description

DYNAMIC N-DIMENSIONAL CUBES FOR HOSTED ANALYTICS}

Cross-Reference to the Related Application

This application is directed to U.S. Provisional Patent Application No. 62 / 015,312, filed June 20, 2014 (named “DYNAMIC CUBES FOR CLOUD-BASED ANALYTICS”); US patent application Ser. No. 14 / 494,506, filed Sep. 23, 2014 (named “DYNAMIC N-DIMENSIONAL CUBES FOR HOSTED ANALYTICS”); US patent application Ser. No. 14 / 494,513, filed Sep. 23, 2014 (named “USE OF DEPENDENCY GRAPHS TO DYNAMICALLY UPDATE N-DIMENSIONAL CUBES”); And US Patent Application No. 14 / 494,524, filed September 23, 2014 ("DATA INTEREST ESTIMATION FOR N-DIMENSIONAL CUBE COMPUTATIONS"), the contents of which are hereby incorporated by reference in their entirety. .

The present application is filed on June 20, 2014, as well as co-pending U.S. Provisional Patent Application No. 62 / 015,302, filed June 20, 2014 (named "EMBEDDABLE CLOUD ANALYTICS" of the invention). Provisional Patent Application No. 62 / 015,308 (named "AUTOMATED HIERARCHY DETECTION FOR CLOUD-BASED ANALYTICS"), and U.S. Provisional Patent Application No. 62 / 015,294 filed June 20, 2014 ("REAL- Also related to TIME HOSTED SYSTEM ANALYTICS "and also to U.S. Provisional Patent Application No. 62 / 015,307 filed on June 20, 2014, entitled" CLOUD ANALYTICS MARKETPLACE ", which is hereby incorporated by reference in its entirety. It is incorporated here.

Data warehouse and online analytics processing (“OLAP”) systems can be used to perform various functions related to data mining, reporting, and forecasting. OLAP systems may enable multidimensional analysis of data typically obtained from transactional systems such as relational databases and loaded into multidimensional cube structures. Data points, such as various aggregate values, can be calculated at each intersection of the various dimensions it contains within the n-dimensional cube structure. Thus, the process of populating a multidimensional cube structure can involve a significant amount of computation. In addition, the n-dimensional cube can be updated on a periodic basis to incorporate new data. Updating the n-dimensional cube may involve recomputing a data point at each intersection of that dimension. Recomputing can be even more burdensome when new dimensions need to be added to the n-dimensional cube. Thus, these types of n-dimensional cube structures are not very suitable for dynamic data environments.

The following detailed description may be better understood when read in conjunction with the accompanying drawings. For purposes of illustration, various examples of aspects of the invention are shown in the drawings; However, the invention is not limited to the specific methods and means disclosed.
1A is a block diagram depicting one embodiment of a system for maintaining an n-dimensional cube usable in association with a cloud-based analytics system;
1B is a block diagram depicting an operation related to the addition of a new dimension observed in a real-time data stream, as an operation of one embodiment of a system for maintaining an n-dimensional cube usable in conjunction with a cloud-based analysis system;
2A is a block diagram depicting an operation of one embodiment of a system for performing a cloud-based analysis system, the operation involving the maintenance of a dependency graph upon addition of a new dimension to an n-dimensional cube, FIG.
2B is a block diagram depicting operations involving the addition of attributes to an n-dimensional cube as operations in one embodiment of a system for performing a cloud-based analysis system;
2C is a diagram depicting one embodiment of a scalable storage mechanism for hierarchical data contained within a slice;
3A is a flowchart depicting one embodiment of a process for maintaining an n-dimensional cube adapted to a cloud-based analytics system;
3B is a flowchart depicting one embodiment of a process for deferring the calculation of data points in an n-dimensional cube;
4 is a flowchart depicting one embodiment of a process for performing deferred computing in an n-dimensional cube adapted to a cloud-based analysis system;
5A is a block diagram depicting one embodiment of a system for providing a hosted analytics system service;
5B is a depiction of a process for the inflow and processing of data from a real time data source;
6 is a block diagram depicting an example of a computing environment in which aspects of the invention may be practiced, and
7 is a block diagram depicting one embodiment of a computing system in which aspects of the invention may be practiced.

Aspects of the invention may be employed to maintain n-dimensional cubes using structures suitable for dynamic data environments, including hosted analytical platform. Embodiments may be employed to provide an analytical system with streams of data that may introduce new attributes, dimensions, or hierarchies, which may be useful for analytical systems when included in an n-dimensional cube. Embodiments may employ n-dimensional cubes structured using slice-based partitioning techniques. Each slice can include a fixed set of dimension values across the slice and data points corresponding to the set of dimension values that are allowed to vary. Attributes, dimensions or hierarchies can be added to the n-dimensional cube by adding new slices or modifying existing slices. The view of the n-dimensional cube may be partially precomputed by forming dependency links between slices, slice regions, and individual data points. The approach described herein can be employed to enable an n-dimensional cube structure that can be expanded or shrunk for additional dimensions and attributes without requiring massive full recalculation or recomputing of the n-dimensional cube.

1A is a block diagram depicting one embodiment of a system for maintaining an n-dimensional cube 100 usable in conjunction with a cloud-based analysis system. A user of a cloud-based analytics system can view the analytics data as a multidimensional array containing aggregate data and related attributes at the intersection point. Multidimensional arrays can be sparse, meaning that a relatively small number of intersection points are associated with the data.

The cloud-based analytical system, although conceptually seen by the user as a multidimensional array, may include an n-dimensional cube 100, which may include a plurality of slices 102-106 and a repository 108. ) May be included. Slice 102 may include an aggregate data set and a set of attributes for a one-dimensional list of array intersections fixed to the remaining dimensions of the n-dimensional cube. In various embodiments, multidimensional “slices” may be employed, ie, slices may also include multidimensional structures that are fixed to the remaining dimensions of the n-dimensional cube.

Repository 108 may include a data repository, such as a relational or non-relational database, that maintains a collection of information about slices 102-106. Non-limiting examples of data that may be maintained in repository 108 include slice identifiers, fixed dimension identifiers, variable dimension identifiers, dependency information, refresh or banality data, and the like.

Slices 102-106 may be maintained on one or more partitions, such as partitions 110, 112. The partition may include a server or other computing node that hosts a database management system or other mechanism for maintaining configuration data of one or more of the slices 102-106. Various partitioning techniques can be employed to distribute the workload and storage requirements between partitions 110 and 112.

Embodiments may also perform replication of partitions 110 and 112 to replicas 114 and 116. Embodiments can use replication to improve the reliability and performance of cloud-based analytical systems employing n-dimensional cube 100. Portions of the n-dimensional cube, such as slices, hierarchies, or regions of the hierarchy, may be partitioned or replicated to meet the computational demands associated with maintaining data points in the n-dimensional cube. For example, the portion of the n-dimensional cube that will be subjected to high read activity may be out-scaled to include a computing node hosting a recordable partition and several additional computing nodes hosting a read-only partition. The portion of the n-dimensional cube associated with the high recording activity may be further partitioned into sub-parts and partitioned among several computing nodes.

In one embodiment, data may be allocated between nodes based on the access pattern. Allocation may involve identifying and implementing partitioning techniques, enabling replication when requested, configuring load balancing mechanisms, and the like. Partitioning and replication may occur at various levels, such as by slices, by hierarchies, by regions of the hierarchy, and the like, as described herein. Access patterns may involve patterns and trends of queries performed on particular slices or by n-dimensional cubes. Various statistics and other metrics regarding access patterns can be collected and used for assignment. These include metrics that record the frequency and rate of update operations, the frequency and rate of operations requiring calculation, the frequency and rate of operations involving data movement, and the like.

In one embodiment, the computational demand associated with the slice portion is compared to the performance characteristics of the partition host, and thus data can be allocated between partition hosts. For example, a slice participating in a high degree of data movement may be placed on the same computing node, placed on a computing node connected to the same branch of the network, or on a computing node connected by a higher-speed network. Can be placed in. Another example involves assigning data that involves a high degree of computing on a computing node where the performance characteristic includes an optimized amount of CPU capability and / or memory.

In one embodiment, data may be allocated between nodes based on security considerations. It may be true that some dimension or level of the hierarchy is used for computing but may not be visible to the user of the n-dimensional cube. Data may be allocated between computing nodes based on security attributes associated with each computing node. For example, the computing node may be configured to process a request originated by the computing node hosting a portion of the n-dimensional cube (so that computing may be performed), and originating from a customer of the n-dimensional cube. It may be further pre-configured to not respond to requests issued by other parties, such as those. Such computing nodes may be assigned data that is required for computing but may not be visible to the user. Other computing nodes that are not so configured can host data that can be viewed by the user.

1B is a block diagram depicting the operation of one embodiment of a system for maintaining an n-dimensional cube usable in connection with a cloud-based analysis system, where the operation involves the addition of a new dimension observed in a real-time data stream. The data stream 160 may include transactional data, real time data, log information, and the like. Data arriving from data stream 160 may include attributes and dimensions that are not included in n-dimensional cube 150 prior to arrival of data.

Embodiments may respond to the arrival of a new dimension 162 by enabling an analysis system that reflects the new dimension 162 without performing destructive operations on the existing n-dimensional cube. Prior art embodiments of n-dimensional cubes may incorporate new dimensions without performing substantial recomputing of existing data points within the n-dimensional cube, such as totals at each intersection, or various derived values. It may not be possible.

As depicted in FIG. 1B, n-dimensional cube 150 may include slices 152-156, where each slice is fixed to a subset of dimensions of the n-dimensional cube and varies across the remaining dimensions. to be. Repository 158 may maintain information about slices 152-156, such as location, last refresh time, dependency information, and the like.

The new slice 164 may be added to the n-dimensional cube 150 in response to the arrival of the new dimension 162 via the data stream 160. The new slice 164 including the new dimension 162 may vary in one or more dimensions. An entry describing the new slice 164 may be added to the repository 158. Existing slices 152-156 may remain in use without the substantial recomputing that occurs prior to their use. Embodiments may add information representing the new slice 164 to repository 158. Embodiments may also add information indicating dependency relationships between existing slices 152-156 and data in new slices 164. The dependency information may include a dependency graph of the data maintained in the slice. The dependency graph may represent dependencies between slices, dimensions, slice regions, data points, and the like. Embodiments may also add information to repository 158 that indicates computational priorities for slices, slice regions, data points within slices, and the like. The priority may be based on various factors, including the estimated probability of interest at the data point, the degree to which other data depends on the calculation, and the like.

 2A is a block diagram depicting an operation of one embodiment of a system for performing a cloud-based analysis system, which involves maintaining a dependency graph upon addition of a new dimension to an n-dimensional cube. FIG. 2A depicts the attendance of an example of an n-dimensional cube initially including a national dimension and a state dimension. One or more measures or other values or attributes may be associated at each intersection of these dimensions. Slice 218 may be fixed to "A" at the country level, as indicated by fixed dimension ("country A") 222. Slice 218 may further include one or more entries along non-fixed dimensions, as depicted in FIG. 2A by variable dimension entries 224-230. Various data points may be associated with dimension entries 224-230. For example, variable dimension entry 226 may be associated with data point 238. Data point 238 may represent various measures, such as total sales in "State B" of "Country A". Although only data point 238 is specifically called out in FIG. 2A, each of the other variable dimension entries 224, 228, 230 may have a similar data point. Similarly, slice 220 may be fixed to country "B" at the country level, as indicated by fixed dimension ("country B") 240, and have variable dimension entries 232-236. Each of which may have an associated data point.

Embodiments can process data stream 200 to incorporate new data into n-dimensional cubes. Incoming data can be associated with existing dimensions and incorporated by marking associated slices, slice regions, data points, etc., as outdated or obsolete. The data dictionary can be used to maintain the banality data for various slices, slice regions, and data points. Embodiments may also determine, via processing of data stream 200, that new dimensions 202 have been encountered and can be incorporated into n-dimensional cubes. For example, FIG. 2A depicts the addition of a new dimension 202, the “region” dimension. The new slice 212 may be added to an n-dimensional cube that already contains slices 218 and 220. The new slice 212 may remain constant at the local dimension with a fixed dimension (“region R”) 204 and one or more other, such as the major dimension, as depicted by the variable dimension entries 208, 210. It can be variable in dimensions.

Embodiments may establish dependency relationships between new slices 212 and existing slices 218 and 220. For example, dependency information 214 may indicate the dependency of information in slice 220 or, more precisely, data point associated with variable dimension entry 208 to variable dimension entry 232. Similarly, dependency information 216 may indicate a dependency of a data point associated with variable dimension entry 210 for slice 218 or variable dimension entry 226. Embodiments may use various levels of granularity and directionality in establishing dependency relationships. For example, dependency relationships may be formed between slices, variable dimension entries, fixed data points, and the like, and may be formed in either direction between existing slices or between an existing slice and a new slice. Embodiments can use dependency relationships to mark slices, slice regions, data points, and the like as stale. Embodiments can also use dependency relationships to locate other components, such as pre-computed totals or calculated additional totals, various derived values, and the like. Note that the variable dimension entries 208, 210 may be formed blank or banal so that the calculation of the associated value may be postponed.

2B is a block diagram depicting an operation of one embodiment of a system for performing a cloud-based analysis system, in which the operation involves adding an attribute to an n-dimensional cube. Embodiments process data stream 250 and determine the presence of new attributes associated with existing elements of slice 258, where slice 258 is a fixed dimension (" Country B ") 260 and variable dimension entry 262. -264). An attribute can be thought of at a conceptual level as a value associated with the intersection of a dimension in an n-dimensional cube. An embodiment may store an attribute using the mechanism depicted in FIG. 2B, associated with one or more attributes, such as attribute entry 256, through a linked list or other structure. Upon detecting a new attribute 252 associated with "Note A", an embodiment can locate the variable dimension entry ("Note A") 262 and add the new attribute 254 to the list of associated attributes. . Embodiments may repeat this operation a number of times for additional slices and fixed dimension entries to which attributes are to be associated.

Dependency information can be identified and stored for an attribute. For example, FIG. 2B depicts dependency information 266. The depicted dependency information 266 may represent a dependency relationship between the new attribute entry 254 and the hierarchy 268. Dependency relationships can also be maintained between existing attributes and hierarchies.

Hierarchies and various derived or calculated values may depend on various attribute values or attribute types. For example, the hierarchy may include aggregate values for sales of the product, filtered by attributes such as color. Changing the value of an attribute may require recomputing the hierarchy. Therefore, there may be a dependency relationship between the product color attribute and the hierarchy. Similar relationships can exist for other derived or calculated values. Newly added attributes can also have relationships with hierarchies. An example may occur when a newly added property is of the same class or similar nature as an existing property. If an existing attribute is an element of an existing hierarchy, the new attribute can be an element of the new hierarchy that is parallel to the existing hierarchy. Rather than calculating the new hierarchy immediately, dependency information can be stored to indicate the relationship between the new attribute and the new hierarchy, thereby enabling deferred or on-demand computing of the new hierarchy sequentially.

2C is a diagram depicting one embodiment of a scalable storage mechanism for hierarchical data contained within a slice. In some cases and embodiments, there may be partitioning of n-dimensional cubes between slices so that each slice of data may be maintained on a separate computing node. In some cases scalability may be achieved by performing replication and load balancing between slices. In other embodiments, partitioning may be done within a slice in addition to or instead of slice-based partitioning.

In some embodiments, the hierarchy of data points contained within a slice may be broken down by regions and stored on various computing nodes. The area of the hierarchy of data points may be referred to as a subset of the hierarchy. The scaling mechanism may be selected for each region (or subset) of the hierarchy based on the computational demand associated with the data point or data points contained within the region.

Slices of n-dimensional cubes may contain various hierarchies of dimensional data. For example, the slice may include sales data aggregated by time unit. For illustrative purposes, FIG. 2C will be described with respect to the time dimension and the sales dimension. However, it will be appreciated that the use of time and sales to illustrate the various aspects of FIG. 2C should not be seen as limiting the scope of the invention.

In FIG. 2C, hierarchy 281 may include hierarchy nodes 286-298. Each node in the hierarchy may be a stored representation of a sum of values. Using the time and sales dimensions as an example, hierarchy nodes 292, 294, 296, and 298 may each contain a sum of sales figures for a six-hour time period. Hierarchy nodes 288 and 290 may each contain a total of six-hour figures. For example, hierarchy node 288 may represent a 12-hour period and contain a total of values associated with hierarchy nodes 292 and 294. Similarly, hierarchy node 290 may represent a second 12-hour period and include a total of values associated with hierarchy nodes 296 and 298. Hierarchy node 286 may include a total for a 24-hour period that includes hierarchy nodes 288 and 290. Embodiments can infer from the inclusion of the time dimension in the hierarchy that more current time periods are more likely to involve frequent recording. Embodiments can use the mapping from the time dimension to the predicted level of updates, thereby estimating the likely computational demands that will accompany data points maintained in the region of the hierarchy.

The scalability of the n-dimensional cube can be increased using a tree-based storage mechanism. In one embodiment, the tree-based storage mechanism may parallelize the hierarchical tree, such as hierarchy 281 in FIG. 2C. The update to the hierarchy, in some embodiments, proceeds as follows: 1) New data is stored in the leaf node and any total value in the leaf can be adjusted; 2) the total value of the parent of the leaf node can be adjusted; And 3) the total value of the parent of the leaf node's parent can be adjusted, and so on.

When using time and sales as an example, the node representing the current time period may be updated frequently. This in turn can cause the ancestors in the n-dimensional cube to be updated frequently, as adjustments to the aggregate values flow up through the chain of inheritance. For example, hierarchy node 298 may represent a current six-time window. When sales data for the current window is collected, the value associated with hierarchy node 298 may be adjusted frequently. This may in turn cause hierarchy nodes 290 and 286 to be adjusted. Some embodiments may postpone grand total calculation at various levels of the hierarchy.

The scaling mechanism for the hierarchical data may be based on the classification of regions in the hierarchy. The classification may include factors such as computational demands imposed on computing nodes hosting some or all of the hierarchy. The classification may include the frequency of activities and the types of activities. For example, area 284 associated with high recording activity may, of course, be associated with hierarchy node 298, although in various cases and embodiments, more than one hierarchy node may be classified in this manner. Can be. The high number of records may be a result of the type of data included in the hierarchy, such as hierarchy node 298 containing data from the current time period. Other areas of hierarchy 281 may be classified as areas 280 associated with high computational load. This area 280 may be associated with more demand associated with calculating the aggregate value. For example, if hierarchy node 298 is being updated to include additional data, its ancestor nodes 290, 286 may be involved in the relatively frequent recalculation of totals or other derived values. Another classification may identify region 282 associated with low recording activity. Additional classification may involve areas involved in frequent query and search operations, such as areas associated with high read activity 283. Classification may also involve those areas with relatively little activity.

The scaling mechanism for maintaining the hierarchy may be based on one or more of the foregoing classifications of computational demands. In one embodiment, hierarchical nodes in region 280 associated with high computational load may be partitioned by further subdividing the computation associated with nodes in that region. For example, the calculations associated with hierarchy node 286 may be performed on a computing node separate from those associated with hierarchy node 290. Calculations associated with hierarchy node 286 may be further partitioned among the various computing nodes. For example, calculations associated with branches of a hierarchy starting with hierarchy node 288 may be performed on a computing node separate from those associated with branches starting with hierarchy node 290.

Hierarchy nodes in the area associated with high recording activity may be partitioned horizontally to distribute the recording load across multiple computing nodes. For areas with low write activity but high read activity, replication may be used to distribute the read load across multiple computing nodes.

Embodiments may also emphasize the use of specific resource types based on the foregoing classification. For example, a computing node maintaining a hierarchy associated with frequent computing or writing may keep data in-memory while those associated with low activity may utilize idiomatic storage.

In various cases and embodiments, regions of the hierarchy may consist of paths through the hierarchy. For example, the region may include a first node, a parent of the first node, and the like. The path in the hierarchy can be maintained in main system memory, or on another relatively low-latency storage device, while the data in the path is frequently accessed. Write operations performed on lower-level nodes in the path may trigger cascading updates. Embodiments may maintain the parent of frequently written nodes in main system memory to efficiently process these and similar types of updates. When the access frequency, especially the write frequency, is above the threshold level, the path can be maintained in main memory. When the access frequency drops below a certain level, the path can be maintained on devices with relatively high latency.

In various embodiments, regions of the hierarchy may be mapped to computing nodes based on the classification of the computing nodes. The hierarchy can be hosted on several computing nodes with potentially variable configurations. Some of the computing nodes may be configured as compute-intensive nodes, which may indicate, for example, that the computing nodes are configured to provide improved efficiency in performing computations. Other computing nodes can be configured to provide improved efficiency for storing data.

 3A is a flowchart depicting one embodiment of a process for maintaining an n-dimensional cube adapted to a cloud-based analysis system. Although described as a sequence of operations, one of ordinary skill in the art should not interpret the described order as limiting the scope of the present invention, and at least some of the described operations may be modified, omitted, reordered, supplemented with additional operations, or parallelized. Will be appreciated. Embodiments of the depicted process may be implemented using various combinations of computer-executable instructions executed by a computing system, such as the computing system described herein.

Operation 300 depicts identifying a new dimension, attribute, measure, or other value that can be incorporated into an n-dimensional cube. Embodiments can process data streams for new data indicating adding new dimensions, attributes, measures, or other values. The data stream may correspond to a real time data source, log file, or other data source that is typically associated with a continuous or semi-continuous stream of data. These data sources can generally be described as providing data on an incremental basis.

The data stream may also be associated with historical data, transaction data, and the like, which may be incorporated or updated periodically in the n-dimensional cube rather than on a continuous basis. This type of data source can generally be described as providing bulk load data.

The process for identifying new data for incorporation into an n-dimensional cube can operate similarly for both incrementally loaded data and bulk-loaded data. Operation 302 depicts adding a new slice to an n-dimensional cube based on the newly found dimension. Referring to FIG. 1A, the addition of a new slice includes assigning a partition to host the slice, such as partition 110 in FIG. 1A, replicating the slice to a replica, such as replica 114, and a repository ( 108) may involve updating. In various embodiments, analytical systems incorporating new dimensions may be performed before these steps are completed.

Operation 304 depicts updating one or more slices based on new attributes, measures, or other values identified in data coming from the data stream. Embodiments can update slice data maintained on a partition and trigger the replication of the data.

Embodiments also allow data dictionaries to be updated to reflect the presence of updated data, including marking slices, slice regions, and data points as obsolete if they are rendered outdated due to newly arrived data. Can cause. Operation 306 depicts maintaining the refresh status and dependency information of slices, slice regions, and data points. Embodiments may employ several different levels of granularity for dependency information. One embodiment may maintain coarse granularity only at the data slice level, for example.

Operation 308 depicts incrementally populating a newly added slice and incrementally refreshing an existing slice. Embodiments may add new slices upon discovery of a new dimension of existence, where a relatively small amount of relevant data—one or even zero as few data points—is available. Thus, a slice is created essentially empty and can be populated as data associated with the slice arrives through the data stream.

As depicted by operation 310, various embodiments may be used in a newly added slice by forming a dependency link from a new slice or from a slice region or data point within a new slice to an existing slice, slice region or data point. You can partially materialize the view. The new slice can be considered partially materialized because the availability of the link can tolerate when it is needed for the responsive calculation of data points in the new slice.

As depicted by operation 312, an embodiment may partially compute data points in an added slice based on priorities for computation. Embodiments may use various factors for determining priorities. In one embodiment, user interest may be estimated by various factors for prioritizing calculations. User interest can be estimated by monitoring mouse movements, such as hovering over data points, for example. The client application can monitor mouse movements and send corresponding information to the cloud-based analytics platform. The information may indicate the area of slice data that the user was hovering over with the mouse, which may indicate a drill-down desire to the data. Embodiments may then trigger the calculation of data needed for drill-down. One embodiment may also estimate interest by classifying the data to be prioritized and correlating the categories to the estimated levels of interest for each category. Various additional techniques may be employed to determine priorities, such as the degree of dependency with other data.

Operation 314 depicts optimizing the computing of the slice by reusing aggregate data. The calculation of the various data points in the slice may involve total values that may be combined to form a larger number of values, or divided to form a smaller number of values. Embodiments may maintain dependency graph information for use with aggregate reuse.

Operation 316 depicts optimizing slice computing based on reflection. Here, the term reflection refers to processing an n-dimensional cube matrix (which can be projected onto one or more slices) on a diagonal axis formed between related dimensions, and the other half using completed computing on one half of the diagonal. It may refer to a technique that involves completing computing on the screen. For example, calculations involving model-year-sales can be reused to perform calculations involving year-model-sales. This technique can be applied in response to having a single key performance indicator (such as sales) distributed over a distribution of attributes whose number is above a threshold.

In various embodiments, new dimensions can be added to the n-dimensional cube. The new dimension can be added in response to various events or conditions, such as receiving a request to add a new dimension, receiving data from a data stream corresponding to a dimension that is not already represented in the n-dimensional cube. have. Embodiments may add new dimensions by forming a data slice and adding it to a plurality of additional data slices that may constitute an n-dimensional cube. Information describing the new data slice, which may include information about the computing node on which the slice is hosted, may be added to the repository containing information about the n-dimensional cube. The repository may include inter-slice dependency information.

The data slice may include a plurality of data points corresponding to the intersections of the new dimension with one or more other dimensions already represented in the n-dimensional cube. Values such as totals and other derived values can be associated with the data points.

Forming a new data slice may include partially materializing a hierarchy of data points in the n-dimensional cube. The partially materialized hierarchy may include calculating zero or more of the values associated with the data points in the hierarchy. The calculation of these data points can be postponed until they are needed. Instead of pre-computing each of the data points, embodiments can pre-calculate the dependency information for the data points. For example, the value associated with the first data point can be used to calculate the value associated with the second data point. Embodiments may identify this dependency upon the addition of a new dimension to the n-dimensional cube and may also store information describing the dependency. Information describing the dependency may be stored in the repository within or outside the data slice. In some cases, there may be an inter-slice dependency. In such a case, an embodiment can store dependency information in a central repository, rather than on the computing node hosting the data slice.

An embodiment can calculate a value associated with a data point based on the determined priority. The priority for the calculation may indicate a relative order for calculating the value associated with the data point, and may also indicate that the value should not be computed until or until it needs to respond to a request to read the value.

Embodiments can adjust the priority of deferred calculations based on various factors. This may include performing the calculation immediately. Embodiments include, but are not limited to, previous access patterns on the same data or similar data, user actions, and the like, such as data in a hierarchy that may be conceptually similar to the hierarchy containing the data points to be calculated. You can adjust the priority of the arguments. For example, an embodiment may determine that certain types of drill-down, drill-up, or pivot operations are performed in common and the associated calculations may be highly prioritized or performed immediately.

Another factor that can be used to prioritize computation is security. Embodiments may determine priorities for calculating data points based on various security attributes, such as those associated with dimensions, hierarchies, or n-dimensional cubes, for example.

An embodiment may determine to calculate or compute a data point based on sorting values indicative of the determined priority for calculation. For example, an embodiment may assign a priority score to a data point (or an area of a slice or hierarchy associated with the data point) and arrange the data points accordingly. Various techniques may be employed to create a compact, orderable structure that represents a priority value associated with a data point.

Embodiments may employ dependency information to identify a path in a hierarchy of data points that may require recalculation following a change to a value at the base of the path. For example, when a value at the base of the hierarchy is updated, the ancestor may require recalculation. Embodiments may identify paths between ancestors and data points associated with the changed values, and mark data points along that path as outdated for descendants. The calculation of ancestor data points can then be prioritized using the various techniques disclosed herein.

In some cases, the data point may depend on an incomplete set of data. For example, the total value for the current 24-hour period may be incomplete until the 24-hour period elapses. Embodiments can track data points associated with an incomplete data set and adjust computing priorities based at least in part on when the data set can be considered complete. For example, the ancestors of the data points that depend on the incomplete data set may be marked as low priority for recomputing while the data set is incomplete. The priority can then be adjusted up when the dataset is complete.

In one embodiment, data points may be computed based on extrapolating values associated with descendants of data points to be computed. For example, the total value for the current weekly sales figures may be incomplete before the last day of the week. However, values for missing data points may be extrapolated based on, for example, the corresponding day in the previous week.

3B is a flow chart depicting one embodiment of a process for deferring the calculation of data points in an n-dimensional cube. Although described as a sequence of operations, one of ordinary skill in the art should not interpret the described order as limiting the scope of the present invention, and at least some of the described operations may be modified, omitted, reordered, supplemented with additional operations, or parallelized. Will be appreciated. Embodiments of the depicted process may be implemented using various combinations of computer-executable instructions executed by a computing system, such as the computing system described herein.

Operation 350 depicts one embodiment of identifying a dependency between a first data point and a second data point. The dependency may reflect the relationship between the two values, such as the first data point serving as input to the calculation used to derive the value for the second data point. When the first data point is changed, the second data point may need to be recalculated to remain accurate. However, embodiments may defer the calculation of the second data point and schedule the calculation using various techniques and mechanisms, as presented herein.

Operation 352 depicts determining the probability that a second data point will be accessed. Access of a data point may involve its use in computing in connection with other data points. For this purpose, the probability of access can be calculated by various embodiments, using a dependency graph or similar structure. Various other factors may be used by various embodiments to determine the probability that a second data point will be accessed.

In one embodiment, the probability that the second data point is accessed is based at least in part on receiving information indicative of user interaction with an interface representing or suggesting a current or future drill-down, drill-up, or pivot operation. Can be determined. In a more general aspect, the user can interact with the user interface in a manner that indicates increased probability or certainty that the data point will be accessed. These actions may include hovering over a data field, drilling down, drilling up, or hovering over a button indicating that pivoting should be performed, and the like.

Embodiments may also consider data that has been sent to the client application for display to a user. For example, if data at level "N" of the hierarchy is being displayed in the client application, data points at levels "N-1" and "N + 1" may have increased likelihood of access.

Embodiments can use a comparison of the time that can elapse in calculating data points with the cost of calculating the data points. In some cases, customers of hosted data analytics services may indicate a preference for performance versus cost. In such case, one embodiment may aggressively prioritize computing to minimize latency. In other cases, the customer may wish to reduce the cost of using the data analytics service and may indicate a preference for cost savings. Cost savings can be achieved, in some cases, as a compromise with the reduced performance that may result from delaying the calculation.

Embodiments may use access patterns for transactional data sources associated with n-dimensional cubes or for n-dimensional cubes. For example, a previous query performed against an n-dimensional cube or against a transactional data source may represent a specific total or other value with greater importance than others. Data points associated with such totals or other values may have increased probability to be accessed. These data points may therefore be assigned a higher priority for calculation than other data points.

4 is a flow diagram depicting one embodiment of a process for performing deferred computing in an n-dimensional cube adapted to a cloud-based analysis system. Although described as a sequence of operations, one of ordinary skill in the art should not interpret the described order as limiting the scope of the present invention, and at least some of the described operations may be modified, omitted, reordered, supplemented with additional operations, or parallelized. Will be appreciated. Embodiments of the depicted process may be implemented using various combinations of computer-executable instructions executed by a computing system, such as the computing system described herein.

Embodiments may add newly identified attributes and associated aggregate coordinates to the n-dimensional cube by attaching data to existing memory structures, as depicted by operation 400. Embodiments may maintain slice data structures in system memory, such as RAM ("RAM"). Embodiments may further maintain a copy of the slice data structure on a backing partition, which may be replicated to additional partitions.

In various embodiments, the new dimension can be made auto-discoverable by the end user. An embodiment can send information indicative of a new dimension to a client device operating an embeddable analyzer system that can display an indication of the new dimension to an end user. The user's response to the new dimension, such as mouse movements or mouse clicks, can be used to measure the user's interest in the new dimension and adjust priorities for computing data points in the n-dimensional cube.

As depicted by operation 402, an embodiment can attach and replace slices through a centralized data dictionary. The data dictionary may include one or more tables in a database management system. Data dictionaries can be partitioned and replicated for the purpose of providing improved load balancing capabilities and increased reliability. As depicted by operation 404, an embodiment may maintain various index structures that represent regions of an n-dimensional cube, which may be referred to by the term matrix or matrices. Embodiments may also maintain index structures for slices, slice regions, and data points.

Embodiments may partially materialize views of n-dimensional cubes by building a dependency tree, as depicted by operation 406. Building the dependency tree can be performed instead of directly calculating the data points at each coordinate intersection in the n-dimensional cube. Various techniques may be employed to build the dependency tree, such as those depicted by operations 408 and 410.

Operation 408 depicts using a unique hierarchy within the dimensional attribute to build a dependency tree. Embodiments may calculate the total at the finest granularity with a higher priority than the total at the coarsest particle size. The fine-particle totals can then be projected to form coarse-grained totals. Embodiments may postpone the calculation of coarse-grained totals until needed, such as in response to an indication of user interest.

Operation 410 depicts the use of inference, estimation, classification models, and other similar techniques to identify n-dimensional cube structures to which similar dependency trees should be applied. New dimensions, measures, or attributes may have similarities to existing dimensions so that the dependency model can be cloned, with or without additional modifications. An embodiment may, in some cases, identify a unique correspondence between new attributes. For example, a new store can have the same data dependencies as an existing store. If no unique correspondence is found, the nearest neighbor can be found using a technique, such as a classification. The dependency tree of the nearest neighbor can then be found and adjusted as needed for application to a new dimension, measure, or property.

Classification and inference techniques can be applied to the access pattern of the n-dimensional cube to identify the n-dimensional cube structure to be cloned. For example, users of n-dimensional cubes can be classified into groups. The access patterns of users in a group can be analyzed, for example, by determining which n-dimensional cube structure is accessed most frequently, identifying typical drill-down depths, identifying common pivots, and the like. . When building an n-dimensional cube structure for new users within the same group, this information can be reflected in various aspects of the new n-dimensional cube structure, such as computing priority.

5A is a block diagram depicting one embodiment of a system for providing a hosted analytics system service. The hosted analyzer system 500 may be managed by a control plane 502 that coordinates the activities of the various modules of the system.

The image rendering 504 module may provide rendering services for embedded user-interface components, such as graphs and charts. The result set management 506 module may maintain historical information, data caches, and the like related to the results of performing the analysis. The user interface catalog 508 module may maintain a repository of user interface elements for an embedded analysis system, such as images, that may be inserted into the user interface of an application that includes embedded analysis system features. The reporting parameter management 510 module may include a repository of parameters to be used to generate an analysis report, such as time period, geographic region, dimensions to include in the report, desired drill-down levels, and the like.

The aggregate 512 module may perform operations to calculate aggregate values in various dimensions and combinations of dimensions. For example, the total 512 module can calculate monthly, weekly, and daily sales data for a particular store, geographic region, and week.

The derived calculation 514 module may perform secondary calculations based on the aggregate data and other information. The custom calculation 516 module can perform report-specific or user-provided calculations. Custom calculations may be provided, for example, by the application publisher.

The scenario layer 518 module may perform operations related to simulation, projection, or other types of “what-if” scenarios. These may be, for example, custom scenarios provided by the application publisher.

The source and connection parameter catalog 520 may maintain information used to locate and access various information sources. Information for locating the source may include a network address, a uniform resource locator (“URL”), and the like. Information for access may include various forms of credentials, accounts, user names, and the like.

The metadata management 522 module is such as relational data source 528, non-relational data source 530, file-based source 532, streaming source 534, and cloud-based data source 536. It may maintain various forms of metadata and other information used to interface with various data sources. Embodiments may employ metadata from the metadata management 522 module in conjunction with the data transformation 524 module. The data transformation 524 module may perform data transformation and data cleansing operations on incoming data.

The scheduler 526 module can adjust the timing of various activities performed by the hosted analyzer system 500. Coordination may involve scheduling n-dimensional cube reconstruction, scheduling data retrieval, and the like.

Various data sources can be employed. These include non-relational data sources 530 as well as relational data sources 528, such as SQL-based relational database management systems. Various non-relational data sources 530 may include NoSQL database systems, key-value pair databases, object-relational databases, and the like. Various file-based sources 532 can be used, such as document repositories, log files, and the like. The log file may also be processed as streaming data source 534, which may also include other types of data sources from which data may be updated on an ongoing basis. Another example that may be classified as another streaming data source 534 is data generated from a videogame, such as a multi-player video game.

Various types of cloud-based data sources 536 may be used. These may include various web sites or data sources maintained by a provider of hosted analytics services, an application publisher, a user of an application, or a third party.

5B depicts a process for the inflow and processing of data from a real time data source. Data source 560 may be communicatively coupled to adapter 556 and purification pipeline 552. Additional data sources, such as data source 562, may be communicatively coupled to other adapters and pipelines, such as adapter 558 and purification pipeline 554.

The adapter 556 may convert the data from the data source 560 into a format suitable for processing by the purification pipeline 552. The operations performed by the purification pipeline 552 may include performing one or more translations or transformations on the incoming data. Examples include stem finding, primitive finding, and the like. Purification pipeline 552 may be multiplexed. This may involve purging along multiple paths to produce data in a normalized format that matches the normalized format used in each destination n-dimensional cube.

5B depicts the analyzer and storage 550 module. This may refer to various components for performing the analysis system, such as modules 502-526 in FIG. 5A. Purified data coming from the purge pipelines 552, 554 may be processed by the analyzer and storage 550 module. Processing may include operations such as performing totals, performing custom calculations, scenario modeling, and the like. Any calculated or derived value, as well as data from the purification pipelines 552, 554, can be routed and stored in a suitable n-dimensional cube.

Embodiments of the present invention may be employed with many types of database management system (“DBMS”). DBMSs are software and hardware systems for maintaining an organized set of data on which storage and retrieval operations can be performed. In a DBMS, data is typically organized by the association between key values and additional data. The nature of the association may be based on the real-world relationships that exist in the set of data, which may be arbitrary. Various operations can be performed by the DBMS, including data definitions, queries, updates, and management. Some DBMSs provide for interacting with the database using a query language such as Structured Query Language ("SQL"), while others use APIs that include operations such as put and get. Interaction with the database may also be based on various protocols or standards, such as hypertext markup language ("HTML") and extended markup language ("XML"). DBMSs may include various architectural components, such as storage engines, which serve to store data on one or more storage devices, such as solid-state drives.

6 is a diagram depicting an example of a distributed computing environment in which aspects of the present invention may be practiced. Various users 600a operate on any type of computing device 602a to communicate over communications network 604 in a process running on various computing nodes 610a, 610b, 610c in data center 620, Interact with a variety of client applications. Alternatively, client application 602b can communicate without user intervention. The communication network 604 may include any combination of communication technologies, including the Internet, wired and wireless LANs, fiber optic networks, satellite communications, and the like. Any number of networking protocols may be employed.

Communication with processes running on computing nodes 610a, 610b, 610c operating within data center 620 may be provided through gateway 606 and router 608. Numerous other network configurations may also be employed. Although not explicitly depicted in FIG. 6, various authentication mechanisms, web services layers, business objects, or other intermediate layers may be provided to mediate communication with processes running on computing nodes 610a, 610b, 610c. . Some of these middle layers may themselves include processes running on one or more of the computing nodes. Computing nodes 610a, 610b, 610c and the processes running therein may also communicate with one another via router 608. Alternatively, separate communication paths may be employed. In some embodiments, data center 620 may be configured to communicate with additional data centers such that computing nodes and processes running therein may communicate with computing nodes and processes operating within other data centers.

Computing node 610a is depicted as residing on physical hardware that includes one or more processors 616, one or more memories 618, and one or more storage devices 614. The process on computing node 610a may be executed with an operating system or, alternatively, a bare-metal process that directly interacts with a physical resource, such as processor 616, memory 618, or storage device 614. metal process).

Computing nodes 610b, 610c are depicted as operating on virtual machine host 612, which can provide shared access to various physical resources such as physical processors, memory, and storage devices. Any number of virtualization mechanisms can be employed to host the computing nodes.

The various computing nodes depicted in FIG. 6 may be configured to host web services, database management systems, business objects, monitoring and diagnostic functions, and the like. Computing nodes may refer to various types of computing resources, such as personal computers, servers, clustered computing devices, and the like. A computing node may refer to a variety of computing devices, such as, for example, cell phones, smartphones, tablets, embedded devices, and the like. When implemented in hardware form, a computing node is generally associated with one or more processors configured to store computer-readable instructions and one or more processors configured to read and execute instructions. The hardware-based computing node may also include one or more storage devices, network interfaces, communication buses, user interface devices, and the like. Computing nodes also encompass virtualized computing resources, such as virtual machines, virtualized bare-metal environments, and the like, implemented with or without a hypervisor. Virtualization-based computing nodes may have virtualized access to hardware resources as well as non-virtualized access. The computing node may be configured to execute one or more application programs as well as an operating system. In some embodiments, the computing node may also include a bare-metal application program.

In at least some embodiments, a server implementing some or all of one or more of the techniques described herein can include a general purpose computer system configured to include or be configured to access one or more computer-accessible media. 7 depicts a general purpose computer system configured to include or be configured to access one or more computer-accessible media. In the illustrated embodiment, computing device 700 is coupled to system memory 720 via input / output (I / O) interface 730 (here, in singular to processor 710 or plural to process 710). One or more processors (710a, 710b, and / or 710n), which may be referred to as < RTI ID = 0.0 > Computing device 700 further includes a network interface 740 coupled to I / O interface 730.

In various embodiments, computing device 700 includes a uniprocessor including one processor 710, or several processors 710 (eg, two, four, eight, or other suitable numbers). May be a multiprocessor system. Processor 710 may be any suitable processor capable of executing instructions. For example, in various embodiments, processor 610 is a general purpose or embedded processor that implements any of a variety of instruction set architectures (ISAs), such as x86, PowerPC, SPARC, or MIPS ISA, or any other suitable ISA. Can be. In a multiprocessor system, each of the processors 610 may implement, but not necessarily, the same ISA in common.

In some embodiments, graphics processing unit (“GPU”) 712 may participate in providing graphics rendering and / or physical processing capabilities. The GPU may include, for example, a highly parallelized processor architecture that is specialized in graphics computation. In some embodiments, processor 710 and GPU 712 may be implemented as one or more of the same type of device.

System memory 720 may be configured to store data and instructions that processor (s) 610 may access. In various embodiments, system memory 720 may be any suitable memory, such as static RAM ("SRAM"), synchronous dynamic RAM ("SDRAM"), nonvolatile / flash®-type memory, or any other type of memory. Even techniques can be implemented using. In the illustrated embodiment, program instructions and data that implement one or more desired functions, such as those methods, techniques, and data described above, are shown to be stored in system memory 720 as code 725 and data 726. It is.

In one embodiment, I / O interface 730 includes I / O interface between any peripherals in the device, including processor 710, system memory 720, network interface (s) 740, or other peripheral interface. O traffic can be configured to adjust. In some embodiments, I / O interface 730 may perform any necessary protocol, timing, or other data conversion to convert data signals from one component (eg, system memory 720) to another component (eg, For example, it may convert to a format suitable for use by the processor 610. In some embodiments, I / O interface 730 is connected to devices attached through various types of peripheral buses, such as, for example, a peripheral component interconnect (PCI) bus standard or a variant of the universal serial bus (USB) standard. Support may be included. In some embodiments, the functionality of I / O interface 730 may be divided into two or more separate components, such as, for example, a north bridge and a south bridge. In addition, in some embodiments, some or all of the functionality of I / O interface 730, such as the interface to system memory 620, may be incorporated directly into processor 710.

Network interface (s) 740 may be used to exchange data between computing system 700 and other devices or devices 760 attached to the network or networks 750, such as, for example, other computer systems or devices. It may be configured to enable. In various embodiments, network interface (s) 740 may support communication via any suitable wired or wireless general data network, such as, for example, an Ethernet network type. Additionally, network interface (s) 740 may be via a telecommunications / telephone network, such as an analog voice network or a digital fiber optic communications network, through a storage area network, such as a fiber channel SAN (storage area network), or any Any type of network and / or protocol may support communication.

In some embodiments, system memory 720 may be one embodiment of computer-accessible media configured to store program instructions and data as described above to implement embodiments of the corresponding methods and apparatus. However, in other embodiments, program instructions and / or data may be received, transmitted, or stored on various other types of computer-accessible media. Generally speaking, computer-accessible media are magnetic or optical media, eg, memory media such as DVD or CD or non-transitory storage coupled to computing device 700 via I / O interface 730. Media may be included. Non-transitory computer-accessible storage media may also be included in some embodiments of computing device 700 as system memory 720 or other type of memory (eg, SDRAM, DDR SDRAM, RDRAM, SRAM, etc.). ), ROM, or any other volatile or non-volatile media. Moreover, computer-accessible media includes signals or transmission media, such as electrical, electromagnetic or digital signals transmitted over a communications medium, such as a network and / or a wireless link, such as those that may be implemented via network interface 740. can do. Some or all of a number of computing devices, such as those illustrated in FIG. 7, may be used to implement the functionality described in various embodiments; For example, software components running on various other devices and servers can collaborate to provide functionality. In some embodiments, some of the described functionality may be implemented using special purpose computer systems, network devices, or storage devices instead of or in addition to using general purpose computer systems. The term "computing device", as used herein, refers to at least all these types of devices, and is not limited to these types of devices.

Compute nodes, which may also be referred to as computing nodes, are implemented on a wide variety of computing environments, such as tablet computers, personal computers, smartphones, game consoles, commodity-hardware computers, virtual machines, web services, computing clusters, and computing devices. Can be. Any of these computing devices or environments may be described as a compute node or computing node for convenience.

A network set up by an entity, such as a public sector organization or company, to provide one or more web services (such as various types of cloud-based computing or storage) accessible to a set of distributed clients over the Internet and / or other networks. It may be called a network. Such provider networks may host a variety of resource pools, such as a collection of physical and / or virtualized computer servers, storage devices, networking equipment, etc. that are required to implement and distribute the infrastructure and web services provided by the provider network. It can include numerous data centers. The resources may be provided to the client in various embodiments in various units associated with the web service, such as a certain amount of storage capacity for the repository, processing capability for processing, instances, a set of related services, and the like. A virtual computing instance can be run on a specified software stack (e.g., sequentially on a hypervisor) and on a specified software stack (e.g., by specifying the type and number of CPUs, main memory size, etc.) One or more servers with a particular version of the operating system).

Various other types of computing devices may be used alone or in combination to implement the resources of a provider network in various other embodiments, including general purpose or special purpose computer servers, storage devices, network devices, and the like. In some embodiments, the client or user may be provided direct access to the resource instance, for example, by granting the administrator login and password to the user. In another embodiment, the provider network operator may allow a client to specify execution requirements for a specified client application and make it suitable for the application (eg, without requiring the client to access the instance or execution platform directly). Support for server instances, Java ^TM virtual machines (JVMs), general purpose or special purpose operating systems, and various interpreted or compiled programming languages-Ruby, Perl, Python, C, C ++, etc. Or may schedule the execution of an application on behalf of a client on an execution platform (such as a high performance computing platform). Certain execution platforms may utilize one or more resource instances in some implementations; In other implementations, multiple execution platforms can be mapped to a single resource instance.

In many environments, operators of provider networks that implement different types of virtualized computing, storage, and / or other network-accessible functionality can enable customers to reserve or purchase access to resources in various resource acquisition modes. have. The computing resource provider may provide functionality for customers to select and launch the desired computing resources, deploy application components to the computing resources, and maintain applications running in the environment. Additionally, computing resource providers can quickly and easily upscale or downsize the number and type of resources allocated to an application, either manually or through automatic scaling, as the demand for the application or its capacity requirements change. It can provide additional functionality for scaling. Computing resources provided by a computing resource provider may be made available to individual units, which may be referred to as instances. An instance can represent a physical server hardware platform, a virtual machine instance running on a server, or some combination of both. Various types and configurations of instances may be available, including different sizes of resources running different operating systems and / or hypervisors, and with various installed software applications, runtimes, and the like. The instance may be further available in a particular availability zone, for example, representing a logical zone, fault tolerance zone, data center, or other geographic location of the underlying computing hardware. Instances can be copied across or within an availability zone to improve redundancy of the instances, and instances can be migrated within or across specific availability zones. As an example, the latency for client communication with a particular server in an availability zone may be less than the latency for client communication with another server. As such, instances can be migrated from higher latency servers to lower latency servers to improve the overall client experience.

In some embodiments, the provider network may be organized into a plurality of geographic regions, each of which may include one or more availability zones. Availability zones (which may also be referred to as availability containers) in turn may include one or more distinct locations or data centers configured in such a way that resources in a given availability zone may be isolated or isolated from failures in other availability zones. have. That is, a failure in one availability zone may not be expected to result in a failure in any other availability zone. Thus, the availability profile of the resource instance is intended to be independent of the availability profile of the resource instance in another availability zone. Clients may be able to protect their applications from failures in a single location by launching multiple application instances in each availability zone. At the same time, in some implementations inexpensive and low latency network connectivity may be provided between resource instances residing within the same geographic area (and network transmission between resources in the same availability zone may be much faster).

Each of the processes, methods and algorithms described in the preceding sections may be embodied in a code module executed by one or more computers or computer processors, and thereby be fully or partially automated. The code module may be stored on any type of non-transitory computer-readable medium or computer storage device, such as a hard drive, solid state memory, optical disk, or the like. Processes and algorithms may be implemented in part or in whole in application-specific circuits. The results of the disclosed processes and process steps can be stored, either permanently or otherwise, in any type of non-transitory computer storage, such as, for example, volatile or non-volatile storage.

The various features and processes described above can be used independently of one another or can be combined in a variety of ways. All possible combinations and sub-combinations are intended to be within the scope of the present invention. In addition, certain method or process blocks may be omitted in some implementations. The methods and processes described herein are also not limited to any particular sequence, and blocks or states associated therewith may be performed in other suitable sequences. For example, the blocks or states described may be performed in a different order than specifically disclosed, or multiple blocks or states may be combined into a single block or state. The block or state of the example can be performed in series, in parallel, or in some other way. Blocks or states may be added to or removed from the disclosed example embodiments. The systems and components of the examples described herein can be configured differently than described. For example, elements may be added to, removed from, or rearranged relative to the disclosed example embodiments.

It will also be appreciated that various items are illustrated as being stored on or in memory while being used, and that these items or portions thereof may be transferred between memory and other storage devices for memory management and data integrity purposes. Alternatively, in other embodiments, some or all of the software modules and / or systems may be executed in memory on other devices and may communicate with the illustrated computing system via computer-to-computer communication. Furthermore, in some embodiments, some or all of the systems and / or modules may be, but are not limited to, one or more application specific semiconductors (ASICs), standard integrated circuits, controllers (eg, by executing appropriate instructions, and In other ways, such as at least in part, in firmware and / or hardware including microcontrollers and / or embedded controllers), field-programmable gate arrays (FPGAs), complex programmable logic devices (CPLDs), and the like. Can be. Some or all of the modules, systems, and data structures may be placed on a computer-readable medium (eg, software) such as a hard disk, memory, network, or portable media article to be read through a suitable connection or by a suitable device. Instructions or structured data). The system, modules and data structures also occur on a variety of computer-readable transmission media (eg, as part of a carrier wave or other analog or digital propagated signal), including wireless-based and wired / cable-based media. And may take various forms (eg, as part of a single or multiplexed analog signal, or as multiple discrete digital packets or frames). Such computer programs may also take other forms in other embodiments. Thus, the present invention may be practiced with other computer system configurations.

The foregoing may be better understood in view of the following provisions:

1. A system for performing a data analysis system, the system having improved scaling characteristics and the system

A plurality of computing nodes comprising a first one or more computing nodes and a second one or more computing nodes, wherein the plurality of computing nodes, when activated, maintain slices of the n-dimensional cube, the slices maintaining a hierarchy of data points. Wherein the data points comprise a plurality of computing nodes corresponding to intersections of at least one fixed dimension and at least one variable dimension; And

And one or more memories having stored thereon computer-readable instructions, wherein the computer-readable instructions, when executed, cause the system to at least

Maintain a first subset of the hierarchy of data points on the first one or more computing nodes, wherein the data points are based on at least in part the association of the data points of the hierarchy with the first classification of computational demand. Included in;

Maintain a second subset of the hierarchy of data points on the second one or more computing nodes, wherein the data points are based on at least in part the association of the data points of the hierarchy with a second classification of computational demand. Included in; And

Process the request by processing on the first one or more computing nodes a first request to access a data point of the hierarchy, wherein the first one or more computing nodes are configured with a scaling mechanism based at least in part on the first classification of computational demand. System.

2. The system of clause 1, wherein the first classification is based on a high write frequency and the scaling mechanism comprises horizontally partitioning data points in the first subset of the hierarchy between the first one or more computing nodes. .

3. The system of clause 1, wherein the first classification is based on a high read frequency and the scaling mechanism comprises the first one or more computing nodes including at least one read-only copy of a partition that is writable.

4. The system of clause 1, wherein, when executed, causes the system to at least

And at least one memory storing computer-readable instructions for classifying regions of the hierarchy based at least in part on a mapping associated with a hierarchy and one or more computational demands associated with the dimensions.

5. The system of clause 1, wherein, when executed, causes the system to at least

Identify a path in the hierarchy based at least in part on the frequency of access to data points in the path; And

 Further comprising one or more memories storing computer-readable instructions for maintaining data points associated with a path in main memory while the frequency of access is above a threshold level.

6. A computer-implemented method for maintaining n-dimensional cubes,

identifying a first subset of the hierarchy of data points in the n-dimensional cube, wherein the subset is identified based at least in part on computational demand associated with the data points in the first subset of hierarchy ;

Selecting a scaling mechanism based at least in part on computational demand associated with the data point; And

Maintaining a first subset of the hierarchy on at least the first computing node and the second computing node, wherein the first subset of the hierarchy is between at least the first computing node and the second computing node based on the selected scaling mechanism. Maintaining partitioning or replicating.

7. The method of clause 6, wherein the scaling mechanism is selected based at least in part on the higher write frequency relative to the read frequency.

8. In accordance with paragraph 7,

Partitioning data points in the first subset of the hierarchy horizontally between at least the first computing node and the second computing node.

9. The clause of clause 6, wherein the scaling mechanism is selected based at least in part on the high frequency of reading relative to the capacity of the first computing node to respond to a request to read data in the first subset of the hierarchy. Way.

10. The provision of paragraph 9,

Replicating at least a portion of the first subset of the hierarchy on the second computing node, further comprising replicating at least a portion of the first subset of the hierarchy maintained on the first computing node. .

11. In the case of paragraph 6,

Classifying the regions of the hierarchy based at least in part on the mapping between the dimension associated with the hierarchy and the computational demand associated with the dimension.

12. The provision of paragraph 6,

Identifying a path in the hierarchy based at least in part on the frequency of access to data points in the path; And

 Maintaining the data point associated with the path in main memory while the frequency of access is above a threshold level.

13. In the case of paragraph 6,

Selecting a second scaling mechanism for the further subset of the hierarchy based at least in part on the computational demand associated with the data points in the additional subset.

14. A non-transitory computer-readable storage medium, wherein when executed by one or more computing devices, causes the one or more computing devices to at least

Determine to store a hierarchy of data points on at least a first computing node and a second computing node;

Store a first subset of the hierarchy of data points on the first one or more computing nodes, wherein the data points are based on at least in part the association of the data points in the hierarchy with the first classification of computational demand. Included in the set; And

Store a second subset of the hierarchy of data points on the second one or more computing nodes, wherein the data points are based on at least in part the association of the data points in the hierarchy with a second classification of computational demand. A non-transitory computer-readable storage medium having stored thereon instructions for inclusion in a set.

15. The method of clause 14, wherein, when executed by one or more computing devices, causes the one or more computing devices to at least

Determine to store a slice of an n-dimensional cube on a plurality of computing nodes including a first computing node and a second computing node, the slice being a data point corresponding to an intersection of at least one fixed dimension and at least one variable dimension; A non-transitory computer-readable storage medium comprising additional instructions for including a hierarchy of additional data points.

16. The method of clause 14, wherein when executed by one or more computing devices, at least one computing device causes at least

And additional instructions for determining to reposition a portion of the first subset of the hierarchy of data points from the first one or more computing nodes, wherein the determination is based at least in part on monitoring computational demand associated with the data point; Non-transitory Computer-readable Storage Media.

17. The non-transitory computer-readable storage of clause 14, wherein the first classification is based on a high write frequency and the data points of the first subset of the hierarchy are partitioned horizontally between the first one or more computing nodes. media.

18. The non-transitory computer-readable storage medium of clause 14, wherein the first classification is based on a high read frequency and the first one or more computing nodes include a recordable partition and a read-only copy.

19. The method of clause 14, wherein, when executed by one or more computing devices, causes the one or more computing devices to at least

A non-transitory computer-readable storage medium comprising additional instructions for classifying regions of a hierarchy based at least in part on a mapping associated with a hierarchy and one or more computational demands associated with the dimensions.

20. The method of clause 14, wherein when executed by one or more computing devices, cause the one or more computing devices to at least

Identify a path in the hierarchy based at least in part on the frequency of access to data points in the path; And

Non-transitory computer-readable storage medium comprising additional instructions for maintaining a data point associated with a path in main memory while the frequency of access is above a threshold level.

And the above may be better understood in view of the following additional set of provisions:

1. A system for performing on-line analytics processing on data involving a real-time data stream, the system comprising

A plurality of computing nodes holding an n-dimensional cube comprising a plurality of dimensions; And

One or more memories having stored thereon computer readable instructions, wherein the computer readable instructions, when executed by the one or more computing nodes, cause the system to at least

Receive information indicating adding an additional dimension to the n-dimensional cube;

Forming a data slice, the data slice comprising a plurality of data points corresponding to intersections of at least one of the additional dimension and the plurality of dimensions; And

Partially materializing the hierarchy of the plurality of data points, and partially materializing the hierarchy includes identifying a dependency between a first data point of the plurality of data points and a second data point of the plurality of data points, Less than all of the plurality of data points are being computed during partial materialization.

2. The system of clause 1, wherein, when executed by one or more computing nodes, causes the system to at least

And at least one memory having stored computer-readable instructions for storing information indicative of the dependency in a repository maintained on at least one of the plurality of computing nodes.

3. The system of clause 2, wherein the repository includes information indicating a second dependency between the data slice and additional data slices of the n-dimensional cube.

4. The system of clause 1, wherein when executed by one or more computing nodes, causes the system to at least

Calculate a value associated with the first data point of the hierarchy;

Identify a path from the first data point to additional data points in the hierarchy, wherein the additional data points are ancestors of the first data point; And

Further comprising one or more memory storing computer-readable instructions for causing the additional data point to record an indication that it is outdated for the first data point.

5. The system of clause 1, wherein the information indicating adding the additional dimension to the n-dimensional cube includes data corresponding to the additional dimension.

6. The system of clause 1, wherein when executed by one or more computing nodes, causes the system to at least

Further comprising one or more memories storing computer-readable instructions that cause the descendant of additional data points to be associated with an incomplete set of data, at least in part, deferring the calculation of additional data points in the hierarchy.

7. A computer-implemented method of maintaining an n-dimensional cube on a plurality of computing nodes, wherein:

Adding an additional dimension to the n-dimensional cube by forming at least a data slice on at least one of the plurality of computing nodes, wherein the data slice corresponds to the intersection of at least one of the additional dimension and the plurality of dimensions of the n-dimensional cube. Adding a plurality of data points; And

Partially forming a hierarchy of the plurality of data points, wherein partially forming the hierarchy comprises identifying a dependency between a first data point of the plurality of data points and a second data point of the plurality of data points. Computer-implemented method comprising the steps.

8. In accordance with paragraph 7,

Identifying a second dependency between a data slice and another data slice maintained on a second at least one of the plurality of computing nodes; And

Storing information indicative of the dependency in a repository maintained on a third at least one of the plurality of computing nodes.

9. The provisions of paragraph 7 provided that

Calculating a value for a first data point in the hierarchy;

Identifying a path from the first data point to additional data points in the hierarchy, wherein the additional data points are ancestors of the first data point; And

And recording an indication that the additional data point is outdated for the first data point.

10. The provision of paragraph 7,

Receiving information indicating adding the additional dimension to the n-dimensional cube, wherein the information comprises data corresponding to the additional dimension.

11. In the provisions of Article 7,

Delaying the calculation of additional data points in the hierarchy based at least in part on the offspring of additional data points associated with the incomplete set of data.

12. The computer-implemented method of clause 11, wherein the additional data point is associated with a total value representing a time period.

13. The provisions of Article 11,

computing additional data points based at least in part on previous access patterns to data in the n-dimensional cube.

14. In the case of paragraph 7,

Computing additional data points in the hierarchy based at least in part on extrapolating values associated with descendants of the additional data points.

15. The provision of paragraph 7 as follows:

Determining that the hierarchy is outdated for changed attribute values;

Prioritizing the calculation of the first level of the hierarchy for changed attribute values;

Deferring a calculation of a second level of the hierarchy, the second level being dependent on the first level; And

Prioritizing the calculation of the second level of the hierarchy in response to at least one of a drill-down, drill-up, or pivot event.

16. A non-transitory computer-readable storage medium, wherein when executed by one or more computing devices, causes the one or more computing devices to at least

Adding additional dimensions to the n-dimensional cube by forming at least data slices on at least one of the plurality of computing nodes, wherein the data slices correspond to at least one intersection of at least one of the additional dimensions and the plurality of dimensions of the n-dimensional cube; A data point of; And

Partially forming a hierarchy of the plurality of data points, wherein partially forming the hierarchy includes identifying a dependency between a first data point of the plurality of data points and a second data point of the plurality of data points A non-transitory computer-readable storage medium having stored thereon.

17. The method of clause 16, wherein, when executed by one or more computing devices, causes the one or more computing devices to at least

Identify a second dependency between a data slice and another data slice maintained on a second at least one of the plurality of computing nodes; And

Non-transitory computer-readable storage medium comprising additional instructions for storing information indicative of the dependency in a repository maintained on a third at least one of the plurality of computing nodes.

18. The method of clause 16, wherein, when executed by one or more computing devices, causes the one or more computing devices to at least

Calculate a value for a first data point in the hierarchy;

Identify a path from the first data point to additional data points in the hierarchy, wherein the additional data points are ancestors of the first data point; And

Non-transitory computer-readable storage medium comprising additional instructions for causing the additional value associated with the additional data point to record an indication that it is outdated for the first data point.

19. The method of clause 16, wherein, when executed by one or more computing devices, causes the one or more computing devices to at least

Non-transitory computer-readable storage medium comprising additional instructions for receiving information indicating adding the additional dimension to the n-dimensional cube, wherein the information comprises data corresponding to the additional dimension.

20. The method of clause 16, wherein when executed by one or more computing devices, cause the one or more computing devices to at least

A non-transitory computer-readable storage medium comprising additional instructions for deferring the calculation of a value associated with additional data points in the hierarchy based at least in part on the offspring of the additional data points associated with the incomplete set of data. .

21. The method of clause 16, wherein, when executed by one or more computing devices, causes the one or more computing devices to at least

Compute a total value for additional data points in the hierarchy based on the incomplete set of data, wherein the incomplete set of data corresponds to a period of time that has not yet elapsed;

Initially defer the calculation of values associated with the ancestors of additional data points; And

Non-transitory computer-readable storage medium comprising additional instructions for prioritizing the calculation of a value associated with an ancestor of additional data points in response to a passage of time.

22. The method of clause 16, wherein when executed by one or more computing devices, cause the one or more computing devices to at least

Determine that the hierarchy is outdated for changed attribute values;

Prioritizing the calculation of the first level of the hierarchy for changed attribute values; And

Initially defer the calculation of the second level of the hierarchy, the second level dependent on the first level; And

Non-transitory computer-readable storage medium comprising additional instructions for prioritizing the calculation of the second level of the hierarchy in response to at least one of a drill-down, drill-up, or pivot event.

In addition, the foregoing may be better understood in view of these additional provisions:

1. A system for performing on-line analytics processing on data involving a real-time data stream, the system comprising

At least one computing node that maintains an n-dimensional cube comprising a plurality of data points and a plurality of data points corresponding to intersections of at least one subset of the plurality of dimensions, wherein the plurality of data points comprise a first data point and a second data point; One or more computing nodes including data points; And

One or more memories that store computer-readable instructions, wherein the computer-readable instructions, when executed by one or more computing nodes, cause the system to at least

Identify a dependency between the first data point and the second data point, the dependency including a calculation of a second data point based on the first data point;

Determine a priority for calculating the second data point, the priority being based at least in part on information indicative of the likelihood of receiving a request to access the second data point; And

Schedule a calculation of a second data point based at least in part on a change to the first data point and on a priority.

2. The system of clause 1, wherein, when executed by one or more computing nodes, causes the system to at least

Further comprising one or more memories storing computer-readable instructions for computing a prospect to receive a request to access a second data point based at least in part on receiving information indicative of interaction with the user interface, system.

3. The system of clause 1, wherein when executed by one or more computing nodes, causes the system to at least

calculate a likelihood to receive a request to access a second data point based at least in part on an access pattern implied by a query executed against at least one of an n-dimensional cube or a transactional data source associated with the n-dimensional cube And at least one memory having stored thereon computer-readable instructions.

4. The system of clause 1, wherein when executed by one or more computing nodes, causes the system to at least

 One or more memories that store computer-readable instructions for calculating a prospect to receive a request to access a second data point based at least in part on a request to perform at least one of drill-down, drill-up, or pivot Further comprising, the system.

5. The system of clause 1, wherein when executed by one or more computing nodes, causes the system to at least

Further comprising one or more memories storing computer-readable instructions for determining a priority for calculating a second value based at least in part on elapsed time during calculating the second data point.

6. A computer-implemented method for calculating a value associated with a data point of an n-dimensional cube, wherein the method is

Identifying a dependency between the first data point and the second data point;

Determining a priority to calculate a second data point, the priority being determined based at least in part on information indicative of the likelihood of receiving a request to access the second data point; And

Scheduling the calculation of the second data point based at least in part on a change to the first data point and on the priority.

7. The computer-implemented method of clause 6, wherein scheduling the calculation of the second data point comprises aligning the first and second data points based at least in part on the priority.

8. In accordance with paragraph 6,

Calculating a likelihood to receive a request to access a second data point based at least in part on receiving information indicative of interaction with at least one of drill-down, drill-up, or pivot control. , Computer-implemented method.

9. In accordance with paragraph 6,

calculating a likelihood of receiving a request to access a second data point based at least in part on an access pattern implied by a query executed against at least one of an n-dimensional cube or a transactional data source associated with the n-dimensional cube And further comprising a step.

10. The computer-implemented method of clause 9, wherein the access pattern comprises access to a total value.

11. In the case of paragraph 6,

Determining a priority for calculating the second data point based at least in part on the cost of computing the second data point.

12. The provision of paragraph 6,

Calculating a second data point based at least in part on receiving a request to access the second data point.

13. In the case of paragraph 6,

And determining, based at least in part on the first data point being sent to the computing device for display to the user, an increased likelihood of receiving a request to access the second value.

14. A non-transitory computer-readable storage medium, wherein when executed by one or more computing devices, causes the one or more computing devices to at least

identify a dependency between a first data point and a second data point of the n-dimensional cube;

Determine a priority for calculating the second data point, the priority being based at least in part on information indicative of the likelihood of receiving a request to access the second data point; And

A non-transitory computer-readable storage medium having stored thereon instructions for scheduling a calculation of a second data point based at least in part on a change to the first data point and based on a priority.

15. The method of clause 14, wherein, when executed by one or more computing devices, causes the one or more computing devices to at least

Additional instructions for calculating a prospect to receive a request to access a second data point based at least in part on receiving information indicative of performing at least one of a drill-down, drill-up, or pivot operation. , Non-transitory computer-readable storage medium.

16. The method of clause 14, wherein when executed by one or more computing devices, at least one computing device causes at least

calculate a likelihood to receive a request to access a second data point based at least in part on an access pattern implied by a query executed against at least one of an n-dimensional cube or a transactional data source associated with the n-dimensional cube And a non-transitory computer-readable storage medium comprising additional instructions.

17. The non-transitory computer- clause 16, wherein the access pattern comprises a query processed by at least one of an n-dimensional cube or a transactional data source associated with the n-dimensional cube, the query comprising a grand total clause. Readable Storage Media.

18. The method of clause 14, wherein when executed by one or more computing devices, at least one computing device causes at least

Non-transitory computer-readable storage medium comprising additional instructions for determining a priority for calculating the second data point based at least in part on the cost of computing the second data point.

19. The method of clause 14, wherein, when executed by one or more computing devices, causes the one or more computing devices to at least

Non-transitory computer-readable storage medium comprising additional instructions for calculating a second data point based at least in part on receiving a request to access the second data point.

20. The method of clause 14, wherein when executed by one or more computing devices, cause the one or more computing devices to at least

A non-transitory computer- comprising additional instructions that determine, based at least in part on the first value being sent to the computing device for display to the user, an increased likelihood of receiving a request to access the second data point. Readable Storage Media.

21. The method of clause 14, wherein when executed by one or more computing devices, at least one computing device causes at least

A non-transitory computer comprising additional instructions that determine to calculate a second data point based at least in part on the priority, the priority being relative to the priority for calculating another data point in the n-dimensional cube. -Readable storage medium.

22. The method of clause 14, wherein when executed by one or more computing devices, causes the one or more computing devices to at least

A non-transitory computer- comprising additional instructions for determining to calculate a second data point based at least in part on a security attribute associated with at least one of the second data point, dimension, hierarchy, or n-dimensional cube. Readable Storage Media.

In particular, the conditional language used herein, such as “may”, “may”, “may”, “may”, “for example”, etc., is specifically described or otherwise used Unless otherwise understood within the context, it is generally intended to convey certain features, elements, and / or steps, including certain embodiments but not other embodiments. Thus, such conditional language generally requires that features, elements, and / or steps be in some way necessary for one or more embodiments, or one or more embodiments, with or without author input or prompts, It is not intended to imply that the steps necessarily include logic for determining in which particular embodiment to be performed or included. The terms "comprising", "comprising", "having" and the like are synonymous and used inclusively, in an open-ended manner, and do not exclude additional elements, features, steps, actions, and the like. In addition, the term “or”, when used to, for example, link elements of a list, includes, but not exclusively, the term “or” to mean one, some, or all of the elements in the list. Used as

While specific example embodiments have been described, these embodiments have been presented by way of example only, and are not intended to limit the scope of the inventions disclosed herein. Thus, nothing in the above description is intended to imply that any particular feature, characteristic, step, module or block is essential or inevitable. Indeed, the novel methods and systems described herein may be embodied in a variety of other forms; Moreover, various omissions, substitutions, and changes in the form of the methods and systems described herein may be made without departing from the spirit of the invention disclosed herein. The accompanying claims and their equivalents are intended to cover such forms or modifications as would fall within the specific scope and spirit of the inventions disclosed herein.

Claims (15)

A system for performing a data analysis system, the system comprising
A plurality of computing nodes comprising a first one or more computing nodes and a second one or more computing nodes, wherein the plurality of computing nodes, when activated, comprise a slice of an n-dimensional cube. The slice comprises a hierarchy of data points, the data points corresponding to intersections of at least one fixed dimension and at least one variable dimension; And
One or more memories having stored thereon computer-readable instructions, wherein the computer-readable instructions, when executed, cause the system to:
Store a first subset of the hierarchy of data points on the first one or more computing nodes, wherein data points of the hierarchy are associated with a first classification of computational demands. Included in the first subset based at least in part on the associated;
Store a second subset of the hierarchy of data points on the second one or more computing nodes, wherein the data points are based at least in part on the data points of the hierarchy associated with a second classification of computational demand. Is included in the second subset; And
Accessing the hierarchical data points on the first one or more computing nodes, wherein the first one or more computing nodes are configured into a scaling mechanism based at least in part on the first classification of computational demand. , system. The method of claim 1, wherein the first classification is based on a high write frequency and the scaling mechanism includes partitioning data points in the first subset of the hierarchy horizontally between the first one or more computing nodes. System. The system of claim 1, wherein the first classification is based on a high read frequency and the scaling mechanism comprises the first one or more computing nodes include at least one read-only copy of a partition that is writable. The system of claim 1, wherein, when executed, causing the system to at least
And at least one memory storing computer-readable instructions for classifying regions of the hierarchy based at least in part on a mapping between a dimension associated with the hierarchy and one or more computational demands associated with the dimension. The system of claim 1, wherein, when executed, causing the system to at least
Identify the path in the hierarchy based at least in part on the frequency of access to data points in the path; And
And at least one memory having stored computer-readable instructions for maintaining the data point associated with the path in main memory while the frequency of access is above a threshold level. A computer-implemented method for maintaining an n-dimensional cube,
Identifying a first subset of a hierarchy of data points in the n-dimensional cube, wherein the subset is identified based at least in part on computational demands associated with data points in the first subset of the hierarchy. Identifying said being;
Selecting a scaling mechanism based at least in part on the computational demand associated with the data point; And
Storing the first subset of the hierarchy on at least a first computing node and a second computing node, wherein the first subset of the hierarchy is based on at least the scaling mechanism selected. And storing the partitioned or replicated between the second computing node and the second computing node. 7. The computer-implemented method of claim 6, wherein the scaling mechanism is selected based at least in part on a higher write frequency relative to the read frequency. The method of claim 7, wherein
Partitioning data points in the first subset of the hierarchy at least horizontally between at least the first computing node and the second computing node. 7. The method of claim 6, wherein the scaling mechanism is selected based at least in part on a high frequency of reading relative to the capacity of the first computing node to respond to a request to read data in the first subset of the hierarchy. Computer-implemented method for maintaining an n-dimensional cube. The method of claim 9,
Replicating at least a portion of the first subset of the hierarchy on the second computing node, wherein the at least a portion of the first subset of the hierarchy is maintained on the first computing node. Further comprising replicating the computer-implemented method for maintaining an n-dimensional cube. The method of claim 6,
Classifying an area of the hierarchy based at least in part on a mapping between the dimension associated with the hierarchy and the computational demand associated with the dimension. The method of claim 6,
Identifying the path in the hierarchy based at least in part on the frequency of access to data points in the path; And
Maintaining the data point associated with the path in main memory while the frequency of access is above a threshold level. The method of claim 6,
Selecting a second scaling mechanism for the additional subset of the hierarchy based at least in part on the computational demand associated with the data points in the additional subset. How to implement. As a system,
A processor; And
A non-transitory computer-readable storage medium, wherein when executed by one or more computing devices, causes the one or more computing devices to at least
Store a first subset of the hierarchy of data points on one or more first computing nodes, wherein the data points are based at least in part on the association of the data points in the hierarchy with a first classification of computational demand. Is included in the subset; And
Store a second subset of the hierarchy of data points on one or more second computing nodes, wherein the data points are based at least in part on the association of the data points in the hierarchy with a second classification of computational demand. And the non-transitory computer-readable storage medium having stored thereon instructions for inclusion in a subset. 15. The computer program product of claim 14, wherein when executed by the one or more computing devices, cause the one or more computing devices to at least
Determine to store a slice of an n-dimensional cube on a plurality of computing nodes including the first computing node and the second computing node, the slice corresponding to an intersection of at least one fixed dimension and at least one variable dimension. And additional instructions to include the hierarchy of data points and additional data points.

Images